OpenAI Launches GPT-5, Unifying Model Lines and Raising the Benchmark Bar

OpenAI has rolled out GPT-5, a single, state-of-the-art architecture that replaces the previous GPT-4o and o-series models. The model promises markedly faster responses while delivering deeper reasoning, setting the stage for broader enterprise adoption.

GPT-5 arrives amid a record AI spending surge: Big Tech firms are set to invest nearly US$400 billion in compute and model training this fiscal year, intensifying the race for differentiated capabilities. By unifying its model family, OpenAI aims to streamline the user experience, no more model picker, and cement its lead before challengers like Anthropic and Google roll out updates. The launch also supports a rapidly growing revenue base expected to hit US$20 billion by year-end.

Early benchmarks underscore the leap: GPT-5 posts 94.6 % on the AIME-2025 math exam, 74.9 % on SWE-bench Verified coding, and 88.4 % on GPQA science—record scores that outstrip GPT-4o by 30–40 percentage points. OpenAI also reports 45 % fewer factual errors in real-world use and latency that “feels instantaneous,” while an optional “thinking mode” extends reasoning chains on demand.

Watch how quickly enterprises shift coding, analytics and support workloads to GPT-5, and how regulators respond to its newly classified “high-risk” bio-security profile. Developers can start testing through the refreshed API today, and free-tier ChatGPT users will receive access via a phased rollout in the coming weeks